Python Data Wrangling: Marketing Campaign Cleaning

Project Overview

In real-world business environments, raw data is often "dirty"—containing typos, logical inconsistencies, and formatting errors. This project demonstrates advanced Data Wrangling and automation techniques using Python (Pandas & NumPy) to transform a messy marketing dataset into a high-integrity asset ready for ROI analysis and supply chain reporting.

---

Key Technical Highlights (Code Snippets)

1. Data Hygiene & Fuzzy Logic

I resolved categorical inconsistencies (e.g., "Gogle" vs "Google Ads") and handled mixed boolean formats ('Yes', 'Y', '1') using mapping dictionaries to ensure uniform data types.

cleanup_map = {
    'Facebok': 'Facebook',
    'Gogle' : 'Google Ads',
    'Insta_gram' : 'Instagram',
    'E-mail' : 'Email'
}
df['channel'] = df['channel'].replace(cleanup_map)

2. Logical Integrity Audits

I implemented custom masks to identify "impossible" records (Clicks > Impressions) and fixed "Time Travel" errors where the end_date occurred before the start_date using automated offsets.

# Fixing chronological errors using a 30-day corrective offset
time_travel_mask = df['end_date'] < df['start_date']
df.loc[time_travel_mask, 'end_date'] = df.loc[time_travel_mask, 'start_date'] + pd.Timedelta(days=30)

3. Outlier Management (Winsorization)

I calculated the Interquartile Range (IQR) to identify extreme spend outliers and capped them at the upper limit to prevent skewed averages in reporting.

Q1 = df['spend'].quantile(0.25)
Q3 = df['spend'].quantile(0.75)
IQR = Q3 - Q1
upper_limit = Q3 + (3 * IQR)
df.loc[df['spend'] > upper_limit, 'spend'] = upper_limit

4. Feature Extraction (String Parsing)

Used string manipulation to extract the Season attribute from complex campaign naming conventions (e.g., extracting "Summer" from "Q1_Summer_Promo").

# Split the name by underscores and grab the second item (index 1)
df['season'] = df['campaign_name'].str.split('_').str[1]

Why This is Business-Critical

Data Integrity: Prevents "impossible" metrics (like Clicks > Impressions) from reaching stakeholders.

Automation: Reduces manual data-prep time by over 80% through repeatable Python scripts.

Reliability: Ensures that supply chain and marketing KPIs (CTR, CPC, AOV) are based on a "Single Version of Truth".

Project Structure

Cleaning Script: View Python Notebook.

Tech Stack: Python 3.x, Pandas, NumPy.

Author

Kaone Edward

• LinkedIn
• GitHub